[SOLVED] Do these clock speeds look correct for a ryzen 3950x?

[barnyard80]

The 3950x is supposed to have a clock speed of 3.5GHz, turboing up to 4.9GHz.

However, when I run some diagnostics, on my machine (Ubuntu 20.04), the results look a lot lower:

lscpu suggests that the max MHz is 3,500MHz, though I did not think that would be the upper bound:

$ lscpu | grep MHz
CPU MHz:                         2196.398
CPU max MHz:                     3500.0000
CPU min MHz:                     2200.0000

Looking at all the cores, I'm disappoint that they all seem well under 3.5GHz:

on@MeshifyC:~$ cat /proc/cpuinfo | grep "processor\|MH"
processor    : 0
cpu MHz        : 2194.666
processor    : 1
cpu MHz        : 2194.979
processor    : 2
cpu MHz        : 2195.510
processor    : 3
cpu MHz        : 2195.003
processor    : 4
cpu MHz        : 2045.898
processor    : 5
cpu MHz        : 2052.893
processor    : 6
cpu MHz        : 3603.375
processor    : 7
cpu MHz        : 2052.369
processor    : 8
cpu MHz        : 2193.840
processor    : 9
cpu MHz        : 2195.657
processor    : 10
cpu MHz        : 2189.764
processor    : 11
cpu MHz        : 2195.105
processor    : 12
cpu MHz        : 2794.171
processor    : 13
cpu MHz        : 2792.243
processor    : 14
cpu MHz        : 1862.944
processor    : 15
cpu MHz        : 1861.263
processor    : 16
cpu MHz        : 2194.732
processor    : 17
cpu MHz        : 2188.485
processor    : 18
cpu MHz        : 2194.160
processor    : 19
cpu MHz        : 2195.480
processor    : 20
cpu MHz        : 2052.064
processor    : 21
cpu MHz        : 2050.417
processor    : 22
cpu MHz        : 3608.316
processor    : 23
cpu MHz        : 2046.137
processor    : 24
cpu MHz        : 2185.432
processor    : 25
cpu MHz        : 2191.569
processor    : 26
cpu MHz        : 2194.620
processor    : 27
cpu MHz        : 2195.207
processor    : 28
cpu MHz        : 2790.551
processor    : 29
cpu MHz        : 2521.658
processor    : 30
cpu MHz        : 1862.732
processor    : 31
cpu MHz        : 1855.240

 

[Darkbreeze]

Looks like you have a problem.

What are your FULL system specs?

What BIOS version are you running on the motherboard?

DO you have the MOST recent AMD chipset driver, from the AMD website, installed?

What is your CPU cooler model and what does your full cooling system consist of including case model, number of case fans, exact locations and airflow direction for EACH case fan, model of case fans and average ambient temperature in the room where your PC is located?

 

[barnyard80]

Hi Darkbreeze,

Thanks again for all your excellent help!

FRACTAL MESHIFY C BLACK GAMING CASE (Window)
AMD Ryzen 9 3950X 16 Core CPU
ASUS® TUF GAMING B550-PLUS
Memory (RAM)32GB Corsair VENGEANCE DDR4 3200MHz (2 x 16GB)
Graphics Card MSI GT 1030 Passive
1TB SEAGATE BARRACUDA SATA-III 3.5" HDD, 6GB/s, 7200RPM, 64MB CACHE
1TB SAMSUNG 970 EVO PLUS M.2, PCIe NVMe (up to 3500MB/R, 3300MB/W)
CORSAIR 750W RMx SERIES™ MODULAR 80 PLUS® GOLD, ULTRA QUIET
Noctua NH-U14S Ultra Quiet Performance CPU Cooler

BIOS version: Not sure. I can check later if it turns out to be important but am doing a Prime95 test right now.

Chipset drivers: I can only find them for Windows. Are they need for Ubuntu?

I have turned off PBO. I have not attempted any overclocking.

I've been playing around with the system since I posted that, and think that perhaps the speeds from my initial post were low because it was so idle.

I have now been running Prime95 for 30 minutes, "Small FFT", having modified local.text with

CpuSupportsAVX=0
CpuSupportsAVX2=0
CpuSupportsAVX512F=0

llscpu is showing:

$ lscpu | grep MHz
CPU MHz:                         3739.643
CPU max MHz:                     3500.0000
CPU min MHz:                     2200.0000

I have PBO turned off. Is the revised figure of 3.7GHz expected?

Also, over this period the CPU temp has ranged between 64C and 65C. Is that a good temperature for this test?

 

[drivinfast247]

FYI: turbo for 3950x is 4.7ghz

 

[Darkbreeze]

BIOS version is ALWAYS important, especially these days. Often, things won't even work right with specific hardware in use, or at all, regardless of hardware when it comes to certain Windows and driver versions, if you are on an early or older BIOS version.

It is not like the old days where you ONLY updated the BIOS if you had a serious problem. Updating the BIOS these days is frequently as much a regular process as updating chipset or other drivers.

If PBO/PBO2 and XFR2 are disabled, then you're not going to see full boost behavior.

Temperature seems good. But, that is the CPU temperature. VRM and chipset temps could be a factor in throttle or lack of boost behavior too, and often are.

In this case though I think that settings is your problem. I'd reset the BIOS to the default configuration and then check your speeds and temps again. You should be able to run Precision boost overdrive enabled with that cooler, especially if you have sufficient airflow through the case but you have not listed your case fan configuration as asked for.

 

[barnyard80]

Hi Darkbreeze,

Apologies for the lack of detail yesterday.

My BIOS version is 0243. I see the latest version on the website is 0805.

The fan configuration is just the stock configuration. Which is a Dynamic X2 GP-12 in the front middle, as input, which therefore points at the GPU. And a Dynamic X2 GP-12 as rear exhaust.

I just turned on PBO with a 200MHz offset, and running Prime95 sent the CPU MHz to around 4,044 and the CPU temp to 90C in under a minute. I stopped it when that happened as I did not want to damage the chip.

 

[Darkbreeze]

Get rid of the offset, and in Prime95 run the Small FFT option, not "Smallest FFT", just "Small FFT" and also DISABLE the AVX2 option which will then un-gray the option to disable the AVX instruction option as well. Run Prime with both AVX and AVX2 disabled, or, you can do the same with OCCT small data set and all AVX options disabled. See what happens then. If you are able to remain below 85°C, you are probably ok. If you can remain below 80°C, you are perfectly fine, and can THEN determine what an appropriate offset for AVX applications is IF you know you run applications that USE AVX.

If you don't run any applications or games that make significant use of AVX, then there is probably not much reason to worry about the offset, but it can't hurt to configure an offset anyhow in case you happen to run something that you didn't know used it.

If you are not thermally compliant, then the fact is that you almost certainly need to add a Top-rear exhaust fan and at least one addition front intake fan, and potentially, a better CPU cooler because while the U14S is a very good CPU cooler for a single finstack cooler, in fact, it is the exact cooler that I am running on my overclocked 6700k, it is almost certainly not capable enough for the 3950x.

 

[barnyard80]

To keep things simple, I set my fan speeds to max.

Without PBO or offset, my temps in Prime95 were 62C.

With PBO, no offset, my temps went over 85C!

I'm surprised PBO made such a big difference!

Having said that, I've read that Prime95 is a very punishing test of CPU temps, and may or may not be reflective of my actual desired use case. I guess I will just have to wait and see.

 

[Darkbreeze]

Prime95 generally IS more stringent than most people's "use case" but that's kind of the point. If you can't run Prime95 without AVX of any flavor running, on the Small FFT option, and it' s IMPORTANT that you are choosing Small FFT and not Large FFT or Blend modes, then you are lacking adequate cooling somewhere.

Try taking OFF the side panel, and running the same test again with PBO enabled, no offset, AVX DISABLED in Prime95 (It is IMPORTANT that you DISABLE AVX when running the Small FFT test if you have no offset in the BIOS configured, and it's kind of the whole point) and see if you still exceed recommended temps.

Out of curiosity, what, exactly, IS your use case?

 

[barnyard80]

I took off the side panel and ran the test again. The temps went over 85C again, but did seem to take a little longer to get there.

My use case is running Python code for data analysis. The code basically loops over lines of text in files, and then does some data analysis, which doesn't use much (if any) scientific stack (i.e probably won't use AVX). Looping over the lines of text is cheap compared to the data analysis, but when it loads each line, it probably gives the CPU a little break as it waits for the data.

The code is suitable for distributing over multiple cores, which is why I got the 3950x.

At a guess (Google hasn't been my friend in researching this), Prime95 could be harsh on the CPU as it just does calculations and does not really wait for data from memory. That won't be the case in my code. However, it is all work-in-progress and I have yet to transfer the code from my old machine to this new one.

 

[Darkbreeze]

So, in that case, maybe try just running Realbench for about an hour. See what happens there with thermals. It's a lot less demanding AND it presents a more realistic workload using a variety of payloads. Run the stress test, not the benchmark.

That, for the overall thermal compliance verification, considering your use case. I'm still of the opinion that if a system cannot run Small FFT without AVX then the cooling in inadequate for the CPU regardless of what the use case is, but if you are confident that the system will never see anything more demanding that the work load you described then that's going to be your call on that.

 

[barnyard80]

I managed to make it run with the code I am writing.

With PBO, the CPU temps varied between 80C and 85C. The clock speeds were just under 4.1GHz across the 32 cores. Does that 4.1GHz seem like a good number?

 

[Darkbreeze]

Between 80 and 85 degrees with WHAT running? Prime or your code?

 

[barnyard80]

It's between 80 and 85 degrees with my own code, which uses all 32 cores.

 

[Darkbreeze]

So, that's probably not good. Anything higher than 80°C but lower than 85°C should be considered and exception, and only for a steady state full TDP stress test load, because then you know that for normal use it's going to be less than that. When your temp is that high under what will be normal use, it's too high and you are looking at potential for electromigration and VT shift or thermal fatigue, prematurely, if you're running consistently at those temps.

Something needs to be lowered or cooling needs to be improved IMO.

 

[barnyard80]

I just added 2 Noctua NF-A15 PWM fans to the system.

So now the fans consist of:

Top front intake: NFA15
Top middle intake: Fractal Dynamic GP12
Top rear exhaust: NFA15
Rear Exhaust: Fractal Dynamic GP12.

The rest of the specs are here:

FRACTAL MESHIFY C BLACK GAMING CASE (Window)
AMD Ryzen 9 3950X 16 Core CPU
ASUS® TUF GAMING B550-PLUS
Memory (RAM)32GB Corsair VENGEANCE DDR4 3200MHz (2 x 16GB)
Graphics Card MSI GT 1030 Passive
1TB SEAGATE BARRACUDA SATA-III 3.5" HDD, 6GB/s, 7200RPM, 64MB CACHE
1TB SAMSUNG 970 EVO PLUS M.2, PCIe NVMe (up to 3500MB/R, 3300MB/W)
CORSAIR 750W RMx SERIES™ MODULAR 80 PLUS® GOLD, ULTRA QUIET
Noctua NH-U14S Ultra Quiet Performance CPU Cooler


I set the fans all to 100% and then re-ran Prime95 with Small FFT, with PBO on. Unfortunately, the CPU temp hit 90C in under a minute. So it seems as though the 2 extra fans didn't make any difference.

If this makes much/any difference... I live in London, and accuweather.com states that today's high temp was 34C, and the high when I ran the prior test (on 25th July) was 19C. I did not measure temperatures directly and have no thermometer in my room. But as a very crude guess, the ambient temperature in the room could be 10-15C higher today.

But to conclude, the results of this experiment seem disappointing.;{

 

[Darkbreeze]

Then it would seem that your cooler is simply not up to the task of cooling that 3950x. At least not with the instruction sets you are running and with PBO enabled. I think you have sufficient case cooling now but you may want to invest in a more capable CPU cooler. The U14S is good, but it hits a thermal wall even with my 6700k which is only a 4/8 CPU, overclocked to 4.6Ghz. I can imagine it falling far short of acceptable performance with a 16/32 CPU that's capable of pushing at least around 160w, CPU only, with PBO enabled under a full load.

I think your ambient definitely plays a role, but even in places like India where the ambient is very high at times, we've seen plenty of users able to manage thermals on these via air coolers, albeit larger ones than your single finstack U14S.

First though, it might be worth taking the time to try and redo the thermal paste. You may have used too much, or too little. You may have not gotten the backplate fully tightened to the mounting brackets or the mounting brackets might not be fully tightened to the heatsink braces. I'd take off the cooler and with only the backplate and standoffs attached, check to see if there is any slack. Some motherboards might be slightly thinner and might allow a seemingly harmless amount of slack to be present but that can be enough to not get a good mount with high pressure. Sometimes we have to space the backplate further away from the motherboard using fiber washers so that the standoffs on the other side that hold the backplate in place are able to FULLY tighten down, before installing the CPU cooler. If the standoffs are even slightly loose or bottomed out before getting tight, it can result in a poor mounting pressure.

It's also as I said possible you pasted wrong or even forgot to remove the protective plastic sheet covering the base of the heatsink. If the heatsink moved around at all during the initial installation it's also possible you ended up with air bubbles and not a good mount. I'd try again before throwing in the towel. If you need to get another heatsink I'd look at one of these coolers below or a good 280-320mm AIO cooler, or even a custom loop if you are so inclined.


Noctua NH-D14 (Replace stock fans with NF-A14 industrialPPC 2000rpm)
Noctua NH-D15/D15 SE-AM4
Noctua NH-D14 (With original fans)
Thermalright Silver arrow IB-E Extreme
Cryorig R1 Ultimate or Universal
Thermalright Legrand Macho RT
Phanteks PH-TC14PE (BK,BL, OR or RD)
Thermalright Macho X2
Deepcool Assassin III